1. Field of the Invention
The present invention relates to a watermark embedding method, watermark detection method, a watermark embedding apparatus, a watermark detection apparatus, storage medium storing a watermark embedding program, storage medium storing a watermark detection program, a watermark system, and a integrated circuit, which, in order to actualize copyright protection for digital content, change personal information content by a small amount that cannot be perceived, and embed watermark so that it cannot be perceived by person, into said content.
2. Description of the Related Art
In recent years, the transfer of digital content via networks has progressed and been facilitated by technologies (such as the JPEG [‘Joint Picture Coding Expert Group’] method for digital information such as voice, moving and still pictures. In addition, the distribution of content via networks has also begun to grow through information infrastructures such as the Internet.
However, content that is distributed via networks is digital information and so it is easy to make perfect copies of personal content. In addition, with information infrastructures such as the Internet, all information users can become information transmitters and so there is copyright infringement through the making and selling of illegal copies. Furthermore, the increase in the processing performance of personal computers facilitates the processing and editing of image and voice data on personal computers, facilitating copyright infringement such as illegal secondary use and use that the authors had not envisaged. The current situation is that the growth of copyright infringement such as this makes authors hesitant about supplying information and thus obstructs the distribution of information via networks.
In addition, separate from this, there is a problem from the point of view of content providers in that “even if one wishes to handle the rights to the content correctly, it is not possible to know where those rights lie if one considers the possibility of incorrect usage” and this also prohibits the reuse of content.
As described below, copyright management/protection systems using conventional technologies have been proposed with respect to these two problems.
Prior art relating to watermark technologies includes A. Piva et. al: “DCT-based Watermark Recovering with Resorting to the Uncorrupted Original Image,” Proceedings of the 1997 International conference on Image Processing (ICIP'97) (Ref. 1 in the following); Takao Nakamura, Hiroshi Ogawa, Youichi Takashima: “A Watermarking Technique for Stile Image,” NTT R&D Vol. 47, No. 6, 1998 (Ref.2 in the following); and Wisetsuit Piyapisuit, Kineo Matsui: “Block Characteristic in Color Image Watermaking Using Equivalent Signal Transform” National Conference of the Institute of Electronics, Information and Communication Engineers, D-11-42, 1999 (Ref.3 in the following).
The method of Ref.1 performs an orthogonal transform of an original image size obtains a frequency component matrix, and changes a coefficient value of a mid-frequency band in the frequency component matrix to embed the watermark. Changing the coefficient value is performed by changing the contrast. This provides robustness with respect to irreversible compressing and color change, and also with respect to changes in the contrast.
In addition, with the method of Ref.1, when the watermark is read, an image which is to be investigated has the watermark read by inverse spread of the coefficient values of the mid-frequency band in the frequency component matrix obtained by orthogonal transform. Through this, it is possible to compare a magnification or reduction with digital content at the original size, align it with the original digital content and perform detection.
In addition, with the method of Ref.1, when reading is performed, an attempt is made using a random number used in the spectral spread and the read watermark is that for which the response is highest. This makes it possible to detect partial images and the like where it is not known where a cut portion lies on the original digital content.
However, the above mentioned Ref.1 involves the following problems.
I. There is a low robustness with respect to partial cutting.
II. There is a large amount of processing involved for an orthogonal transform of an original image at the full size thereof, which causes a local degree of complexity.
III. There is a great amount of image deterioration when the watermark is embedded.
IV. Attempted reading is performed for all values obtained for the watermark when the watermark is detected.
With the method of Ref.2, an image is generally divided into large blocks of a block size used by non-reversible compression, an orthogonal transform is performed for each block and a frequency component matrix is obtained and normalized. A sequence that performs embedding is then selected using a key, and an inverse orthogonal transform is performed to obtain the image having an embedded watermark. With this method, there is robustness with respect to non-reversible compressing but there are problems of color change, contrast change and a low robustness with respect to partial cutting. In addition, although this method aims to improve the image quality/robustness by normalizing and embedding weakly on flat portions and strongly on complex portions of the image, it does not function effectively. Changing the low-frequency band improves the robustness.
When there is watermark detection using Ref.2, processing is performed up to normalization when there is embedding, and the watermark is detected from the key and the normalized frequency component matrix. This means that when the a block size is the same as the image size, there is robustness with respect to magnification and reduction. However, quantitative evaluations have not been performed for the reliability of the detected watermark. In addition, there is also a problem of a large amount of image deterioration when the watermark is embedded.
With the method of Ref.3, processing is implemented to take a transform matrix of an equivalent signal transform (such a RGB<-> YCbCr or the like) for a color image and change it slightly in accordance with watermark for something defined. Watermark information detection for Ref.3 involves performing an equivalent signal transform on the detected object image and then using statistical values to detect whether or not there is a special, status due to the watermark. With this detection, the method of Ref.3 has a robustness with respect to geometrical changes of the image. However, the method is extremely weak with respect to non-reversible compression and color changes. In addition, when there is multi-bit embedding, the embedding is for one bit per block and so there is a problem that the robustness with respect to geometrical transformation is extremely weak.